Architecture Decision Records

docs/adr/0001-record-architecture-decisions.md

@@ -0,0 +1,21 @@
+# 1. Record architecture decisions
+
+Date: 2024-06-09
+
+## Status
+
+Accepted
+
+## Context
+
+We need to record the architectural decisions made in this project to ensure that future contributors understand the reasoning behind key choices and can maintain consistency as the project evolves.
+
+## Decision
+
+We will use Architecture Decision Records (ADRs), following the format described by Michael Nygard in [Documenting Architecture Decisions](https://cognitect.com/blog/2011/11/15/documenting-architecture-decisions.html).
+
+## Consequences
+
+- All significant architectural decisions will be documented in the `adr/` directory.
+- Each ADR will have a unique number and a clear title.
+- ADRs will be kept under version control.

docs/adr/0002-python-monorepo-structure.md

@@ -0,0 +1,27 @@
+# 2. Use a Python monorepo structure
+
+Date: 2024-06-09
+
+## Status
+
+Accepted
+
+## Context
+
+The project consists of multiple components: core logic, document loaders, models, tests, and examples. Managing these as a monorepo simplifies dependency management, testing, and deployment.
+
+## Decision
+
+We will use a single Python repository with the following structure:
+
+- `extract_thinker/` for core code and modules
+- `tests/` for all test code
+- `examples/` for usage examples and scripts
+- `docs/` for documentation
+- Project configuration files at the root
+
+## Consequences
+
+- Easier to manage dependencies and code sharing between modules.
+- Simplifies CI/CD and testing.
+- All code and documentation are versioned together.

docs/adr/0003-document-loader-plugin-architecture.md

@@ -0,0 +1,24 @@
+# 3. Document loader plugin architecture
+
+Date: 2024-06-09
+
+## Status
+
+Accepted
+
+## Context
+
+ExtractThinker needs to support multiple document types and OCR backends (Tesseract, EasyOCR, etc.). We want to make it easy to add new loaders without modifying the core extractor logic.
+
+## Decision
+
+We will use a plugin-like architecture for document loaders:
+- Each loader implements a common interface (`DocumentLoader`).
+- Loaders are registered and discovered by the extractor at runtime.
+- Loader configuration is handled via config classes (e.g., `EasyOCRConfig`).
+
+## Consequences
+
+- New loaders can be added with minimal changes to the core.
+- Loader-specific dependencies are isolated.
+- The extractor can select the appropriate loader based on file type or capabilities.

docs/adr/0004-usepytest-for-testing.md

@@ -0,0 +1,21 @@
+# 4. Use pytest for testing
+
+Date: 2024-06-09
+
+## Status
+
+Accepted
+
+## Context
+
+We need a robust, flexible, and widely adopted testing framework for Python.
+
+## Decision
+
+We will use `pytest` for all unit and integration tests.
+
+## Consequences
+
+- Tests are easy to write and maintain.
+- Fixtures and plugins can be leveraged for complex test scenarios.
+- Test discovery and reporting are standardized.

docs/adr/0005-use-ruff-and-flake8-for-linting.md

@@ -0,0 +1,21 @@
+# 5. Use Ruff and Flake8 for linting
+
+Date: 2024-06-09
+
+## Status
+
+Accepted
+
+## Context
+
+Consistent code style and static analysis are important for maintainability and code quality.
+
+## Decision
+
+We will use [Ruff](https://github.com/astral-sh/ruff) as the primary linter, with Flake8 as a secondary tool for compatibility and additional checks.
+
+## Consequences
+
+- Code style is enforced automatically.
+- Linting errors are caught early in development and CI.
+- Contributors must ensure code passes linting before merging.

docs/adr/0006-llm-agnostic-extraction.md

@@ -0,0 +1,20 @@
+# 6. LLM-agnostic extraction
+
+Date: 2024-06-09
+
+## Status
+
+Accepted
+
+## Context
+
+The project aims to support multiple LLMs (Language Model APIs) for extraction, not just a single provider.
+
+## Decision
+
+The extractor will be designed to load and use any LLM backend, with the LLM set at runtime via configuration.
+
+## Consequences
+
+- The system is flexible and can adapt to new LLMs as they become available.
+- Users can select the LLM that best fits their needs and budget.